OpenAI announces launch ofGPT-5.2-Codexwith advances in agentic programming and cybersecurity. The company makes the model immediately available to paid ChatGPT users on all Codex surfaces. Access via API will occur in the coming weeks. Esse model represents an optimized version of GPT-5.2 specifically for complex encoding tasks in the Codex environment.
GPT-5.2-Codex stands out for improvements in understanding long contexts and native compression. Essas features allow extended sessions on large repositories without loss of efficiency. Ele outperforms predecessors in benchmarks such as SWE-Bench Pro, with 56.4% accuracy, and Terminal-Bench 2.0, reaching 64%.
OpenAI launches an invitational pilot for verified defensive cybersecurity professionals. Esse program offers access to more permissive capabilities with a focus on responsible use.
Improvements in long-horizon tasks
GPT-5.2-Codex incorporates context compression to maintain coherence across extended jobs. Desenvolvedores perform massive refactorings and code migrations with greater reliability. The model better handles plan changes during iterations.
It processes large repositories keeping the context intact over hours. Tarefas that require multiple attempts benefit from improved persistence.
Performance on specialized benchmarks
The model records significant advances in real software engineering assessments.
- SWE-Bench Pro: 56.4% accuracy, outperforming GPT-5.2 by 55.6%.
- Terminal-Bench 2.0: 64% accuracy, up from 62.2% for GPT-5.2.
These tests simulate authentic terminal environments, including code compilation and server configuration. GPT-5.2-Codex demonstrates greater effectiveness in native Windows scenarios.
Advanced capabilities in vision and prototyping
GPT-5.2-Codex accurately interprets shared screenshots and technical diagrams. Ele quickly converts design mocks into functional prototypes.
Developers collaborate iteratively to evolve prototypes into production. Essa functionality accelerates visual workflows in interface projects.
Advances in defensive cybersecurity
The model features stronger cyber capabilities than any previous version of OpenAI. Ele assists in vulnerability analysis and fuzzing in controlled environments.
A researcher used the predecessor GPT-5.1-Codex-Max to identify recently disclosed flaws in React. GPT-5.2-Codex elevates this potential for defensive tasks at scale.
OpenAI implements safeguards for dual-use risks. The model does not reach “High” level at Preparedness Framework.
Careful deployment strategy
The release prioritizes paid ChatGPT users with immediate access on Codex. The extension to the API occurs gradually with security testing.
The invitational pilot selects professionals with a history of responsible disclosure. Organizações focused on cyber defense receive expanded access.
This approach balances innovation with mitigating potential risks.
Integration with existing tools
Simple installation via npm i -g @openai/codex command makes it easy to get started. The model integrates with the Codex CLI for on-premises and cloud workflows.
Developers maintain extended sessions with preserved token efficiency. Chamadas of tools become more reliable in agentic tasks.
Benefits for professional software engineering
GPT-5.2-Codex serves as a dependable partner in complex real-world projects. Ele reduces time for refactoring and feature development.
Companies accelerate engineering teams with autonomous assistance. The focus on factual accuracy minimizes errors in long contexts.
Perspectives on responsible use
OpenAI monitors progress to adjust safeguards in future iterations. Capacidades defensives strengthen legitimate vulnerability searches.
Qualified professionals contribute feedback on the exclusive pilot. Essa collaboration refines the balance between accessibility and security.
GPT-5.2-Codex positions itself as an essential tool for advanced coding. Ele combines cutting-edge performance with accountability in sensitive domains.